Tower Semiconductor

Phase A — Understand the business

Lens 1 · Company Overview

Tower Semiconductor Ltd. (Migdal Haemek, Israel; founded 1993 on the carcass of a National Semiconductor fab) is a pure-play, independent specialty foundry — it manufactures wafers to customers' designs and "does not offer products of its own". It is not a leading-edge logic foundry; it deliberately does not chase TSMC down the node curve. Its geometries are 0.35 / 0.18 / 0.16 / 0.13-micron on 200mm and 65nm on 300mm. The whole company is run as a single reportable segment — "analog foundry operations" (CODM = the CEO). The value proposition is the process menu, not the transistor count: SiGe BiCMOS, silicon photonics (SiPho) incl. silicon-nitride waveguides, RF-SOI antenna switches, CMOS image sensors, power-management (BCD), MEMS and high-voltage CMOS.

• How it makes money: priced per wafer, by process value and capacity utilization; customers place POs 2–6 months ahead. Increasingly it is also taking multi-year capacity-reservation contracts with up-front prepayments in SiPho (a structural shift toward take-or-pay — see Lens 5).
• Customers: fabless analog/mixed-signal players, IDMs, and "module integrators for AI and data centers." Largest single customer = NTCJ (the Nuvoton/ex-Panasonic Japan entity), 11% of FY25 revenue; the next seven customers each ran 4–7%; ~50% from a long tail of smaller names. End-markets: AI/datacenter optical comms, smartphones (RF), automotive, industrial, medical, consumer.
• Competitors: in specialty — GlobalFoundries (RF), Vanguard, DongBu, X-Fab, Hua Hong; at the edges the giants (TSMC, UMC, SMIC) that also sell some specialty capacity.
• Verdict: a high-quality niche operator that, after 30 years as a sub-scale cyclical, has stumbled into being the merchant-foundry pick-and-shovel for AI optical interconnect — the one specialty process (SiPho) the hyperscaler optics build-out cannot easily source elsewhere.

Lens 2 · Supply Chain

Map: III-V epi & substrate suppliers → Tower fabs → fabless/IDM/optical-module integrators → hyperscaler / OEM end-buyer.

• Upstream inputs: raw silicon wafers; Indium-Phosphide (InP) epiwafers — Tower signed a multi-year InP epiwafer supply deal with IQE plc specifically for AI-datacenter photonics (and the same agreement settled all prior IP litigation between them via a worldwide porous-silicon license). SiGe and silicon-nitride deposition materials; semiconductor capital equipment (Applied Materials / ASML-class lithography, etc.).
• The company itself — fab footprint (the asset is geographic + process diversification):
• Fab 1 (Israel, 150mm) — discontinued in 2025, flows consolidated into Fab 2.
• Fab 2 (Migdal Haemek, Israel, 200mm) — SiGe, SiPho, CIS, RF-SOI, power.
• Fab 3 / NPB Co. (Newport Beach, CA, 200mm) — SiPho + analog/mixed-signal. (Note the lease dispute — Lens 10.)
• Fab 9 / Tower SA (San Antonio, TX, 200mm, ex-Maxim) — CMOS/power/SiGe/SiPho.
• TPSCo (Japan, 51%-owned, 200mm + 300mm) — 65/180nm RF, power, CIS. Being restructured (Mar 2026): Tower takes 100% of the 300mm Fab 7; Nuvoton takes the 200mm Fab 5; mutual long-term supply agreements; targeted close Apr 1, 2027.
• Fab 10 / TSIT (Agrate, Italy, 300mm 65nm) — a collaborative cleanroom share with STMicroelectronics; Tower owns ~1/3 of the cleanroom, ST manages the facility; reached volume production in Q4 2024. This is where the new 65nm BCD power platform runs.
• Intel New Mexico "capacity corridor" (300mm) — a Sep-2023 agreement for Intel to build capacity to Tower's instruction; Intel has stated it will not perform and the parties are in mediation. Treat as optionality lost, not as committed capacity.
• Downstream: Tower ships wafers; customers handle dicing/assembly/test. End-demand pull = NVIDIA (Tower is co-developing 1.6T datacenter optical modules for NVIDIA networking protocols, announced Feb 2026).
• Chokepoints / single-source risks: (1) the Agrate 300mm dependency on ST — Tower does not control that facility; (2) InP epi from IQE is a narrow upstream node for the photonics ramp; (3) Israel concentration — substantially all of Tower-parent's long-lived assets are in Israel, a geopolitical single point (war-driven equipment-install delays already flagged).

Lens 3 · Competitive Advantages (moats)

• Process-lock / switching costs (the real moat): "customers that use our specialty process technologies cannot easily transfer designs to another foundry because the analog characteristics of the design are dependent upon the specific process technology" — the PDKs and device models are foundry-specific. Analog designs are co-developed with the process; requalifying elsewhere is multi-quarter and risky. This is a genuine qualitative moat, stronger than in commodity CMOS.
• Scarcity moat in SiPho: the engineering community with specialty-process expertise is "relatively small," and the capital + know-how to offer it has "limited the number of foundries capable". In silicon photonics specifically, Tower is one of very few merchant foundries with a qualified, volume platform — the reason it can sign $1.3B of 2027 contracts and collect prepayments (Lens 5). Demand is expected to outstrip supply through at least 2027.
• Bargaining power — improving but not dominant: historically a price-taker in a cyclical wafer market; the SiPho prepayment/reservation model inverts that for the hot product line (customers now fund Tower's capex up front). But the long tail of analog customers and the single 11% customer mean it is not TSMC — it cannot dictate terms broadly.
• Where the moat is thin: it is not a cost-leader (200mm-heavy, sub-scale vs. GF/Hua Hong); it does not have CoWoS-style advanced packaging; and the moat is IP-contestable — see the GlobalFoundries suit (Lens 13). Brand/marketing are B2B-technical (TGS symposia, PDKs), not a consumer moat.

Lens 4 · Segments

Tower reports one segment (analog foundry), so there is no product-line P&L in the filing — the only structured breakouts are geography and customer concentration. Provenance note: segments.csv in the research layer is empty (headers only); all figures below are ``.

Revenue by geography (by customer HQ, % of total):

The visible mix-shift is toward Asia-ex-Japan and away from Japan/Europe — consistent with the SiPho/RF ramp pulling in Asian fabless optical-module and handset customers, while the legacy Japanese (Panasonic-lineage) volume rolls off.

Product mix (qualitative, from MD&A + Q1'26 call): the FY25 revenue increase was driven by higher wafer shipments incl. the Agrate 300mm facility running a full year. The growth vector is unambiguous from the live data: Silicon Photonics grew ~3x YoY in Q1 2026 and is now the swing factor in mix and margin. Power-management (65nm BCD on 300mm) is the emerging second leg for AI/datacenter power delivery + automotive.

Phase B — Measure performance

Lens 5 · Earnings Result

Two prints matter: the FY2025 annual (audited, on the shelf) and the Q1 2026 print (the inflection).

FY2025 (audited, 20-F):

• Revenue $1,566.1M, +9.0% YoY (FY24 $1,436.1M; FY23 $1,422.7M).
• Gross profit $363.9M (23.2% margin) vs 23.6% FY24 — a slight dilution as Agrate ramp depreciation hit COGS.
• Operating profit $194.2M (12.4%) vs $191.3M (13.3%) FY24.
• Net profit $218.8M; attributable to company $220.5M. EPS $1.97 basic / $1.94 diluted (110–114M shares).
• Comparability flag: FY2023 operating profit of $547.3M and EPS $4.70 are not organic — they include the $313.5M net Intel merger-termination fee. Underlying 2023 ≈ flat with 2024. Always strip this when trend-reading.
• Financing income $56.7M — a real earnings contributor (interest on ~$1.15B cash/deposits + JPY hedging gains). ~24% of FY25 pre-tax profit came from below the operating line; quality-of-earnings caveat (Lens 10).

Q1 2026 (the print that moved the stock):

• Revenue ~$414M, +15% YoY (vs $358M Q1'25).
• Gross profit +52% YoY to ~$111M; operating profit +96% YoY to ~$65M — operating leverage inflecting hard as high-value SiPho fills capacity.
• Net profit $65M, EPS $0.58 basic / $0.57 diluted (adj. EPS ~$0.65 vs consensus, a beat).
• Q2 2026 guide: $455M (company record), +22% YoY / +10% QoQ.
• Balance-sheet tell: $290M of customer prepayments for SiPho landed in Q1 operating cash flow — visible already in the FY25 balance sheet as deferred revenue/advances building, and the "other long-term assets" jump to $141M (FY24 $23M) and the +$106M "other long-term assets" use of cash is the prepayment/capacity-deposit machinery turning over.
• Market reaction: the stock is up roughly 5x from the failed-Intel-deal level (~$45 → ~$250), market cap >$25B. What's priced in: the 2028 model and continued SiPho beats — i.e., a lot.

Lens 6 · Earnings Calls (sentiment trend)

No transcripts on the shelf (transcripts/ empty); this lens is ``.

• Tone trajectory: from cautious-cyclical (2023–24, post-Intel, "macro soft, utilization low") to overtly bullish and capacity-constrained by Q1 2026 — management language shifted to "record," "sequential QoQ revenue and margin growth throughout 2026," and reiterating a hard numerical 2028 target of $2.8B revenue / $750M net profit. The Substack/sell-side read was literally "It's all about silicon photonics now".
• What they started saying: silicon photonics as the headline; contracted revenue and prepayments (a new vocabulary of demand visibility); NVIDIA by name (1.6T modules).
• What they stopped saying: the defensive "we'll manage through the downturn" framing of 2023–24; Intel (now only a mediation footnote).
• Read: the sentiment shift is fact-backed (the contracts and prepayments are disclosed, not hand-wave), which is the right kind of management optimism — but it also means expectations are now maximal, so any SiPho slip will be punished asymmetrically.

Lens 7 · Comps

Specialty-foundry peer table.

Read: on every sourced metric Tower trades at a steep premium to its closest listed peer GF (~66–74x fwd P/E vs GF ~57x trailing; EV/EBITDA ~50x vs the semi-median ~36x fwd P/E). The market is valuing Tower not as a specialty foundry but as a SiPho/AI-optics pure-play with a 2028 hyper-growth model. The 5-yr-avg-ROE column is n/a (would require a clean ROE series; FY25 ROE ≈ $220.5M / ~$2.92B equity ≈ 7.6% — structurally low for the multiple, the central valuation tension).

Lens 8 · Stock-Price Catalysts (>5% moves, last ~5y)

Mostly ``; the pattern is the analytically useful part.

• Feb 2022: Intel agrees to buy Tower for $5.4B / $53 a share → gap up, stock pinned near the deal price for 18 months.
• Aug 16, 2023: Intel terminates (no China SAMR approval by deadline); Tower collects a $353M reverse break fee ($290M net to Q3'23 net profit) → stock de-rated back to the ~$30s–40s as the takeout premium evaporated.
• 2024: Agrate 300mm reaches volume production; grind higher on the analog recovery.
• Feb 5, 2026: NVIDIA partnership for 1.6T optical modules → sharp pop; the re-rating begins.
• ~Q1 2026: $1.3B SiPho contracts for 2027 + $290M prepayments announced with the print → "valuation over $25B".
• Through mid-2026: repeated analyst-target hikes (Benchmark to $335 from $230; Street range $300–$335, avg ~$314).
• Pattern → what the market actually reacts to for TSEM: (1) M&A / strategic ownership events (the Intel saga dominated 2022–23), and now (2) silicon-photonics demand signals — named hyperscaler wins and contracted dollar figures. It does not trade much on the legacy analog cycle anymore. The corollary: the next >5% moves will be set by SiPho contract cadence and any crack in the NVIDIA/optics demand narrative — and by the GF litigation docket.

Phase C — Judge people & books

Lens 9 · Management

• CEO Russell C. Ellwanger (71) — CEO since May 2005 (~20 years); ex-Applied Materials group VP and Novellus/Philips Semiconductors.. Track record: took a single near-bankrupt Israeli fab and built a four-country, $1.5B-revenue specialty-foundry network via serial bolt-ons (Jazz/Newport Beach 2008, TPSCo/Panasonic 2014, Maxim/San Antonio 2016, ST-Agrate JV 2021) — a genuine M&A-driven roll-up operator. He also steered the company through the Intel deal collapse without value destruction (kept the $353M fee, redeployed into SiPho). The current SiPho inflection happened on his watch.
• Bench: Dr. Marco Racanelli (President, since Nov 2023) — deep SiGe/BiCMOS/RF pedigree (Jazz, Conexant, Motorola), 35+ patents; the technical architect of the specialty franchise. Oren Shirazi (CFO since 2004, ~21 yrs) — long-tenured, Israeli CPA. Very low management turnover.
• Skin in the game: thin direct ownership — as of Mar 31 2026 all directors + senior management together beneficially owned ~0.50% of shares plus ~1.15M RSUs/PSUs. Mitigant: the CEO is subject to a minimum-holding requirement of ≥3x base salary in shares, with which he is compliant. Comp is performance-leaning — FY25 grant ~$11.9M total value, 60% PSUs/MSUs tied to net-profit, revenue and 3-year share-price hurdles (the upside PSUs earned only 29% on financial metrics but 100% on the share-price metric in 2025). That is reasonable alignment, though the cash-vs-equity-vs-dilution mix bears watching (SBC ran ~$37.8M in FY25, ~2.4% of revenue).
• Capital allocation: no dividend, no active buyback — 100% of capital is being plowed into the $920M SiPho/SiGe/power capex program. Net cash ~$990M (Lens 10) gives them the firepower. The quality of this allocation is the whole bet: if SiPho demand is durable, reinvesting at high incremental returns is exactly right; if it's a bubble, this is capacity built into a top.
• Archetype: a professional-manager / serial-acquirer operator (not a founder), long-tenured and credible, who has earned the benefit of the doubt — but who has never run the company through a demand-driven hyper-growth phase before (his prior cycles were survival + roll-up).
• Red flags: none egregious. The 0.50% insider ownership is the main governance soft-spot; related-party items are routine; no promotional stock-pumping history.

Lens 10 · Forensic Red Flags

Ground: financials.csv empty → figures ``. Auditor: Brightman Almagor Zohar (Deloitte Israel), opinion dated Mar 2, 2026.

• Earnings quality / financing income: ~$56.7M of FY25 financing income = ~24% of pre-tax profit. It is real (interest on ~$1.15B cash + deposits, plus JPY hedge gains), but it is not operating and is rate-sensitive — non-GAAP "core" margins are thinner than headline net margin implies. Watch if rates fall.
• Cash vs earnings: clean — FY25 operating cash flow $395.5M comfortably exceeds net profit $218.8M (D&A $303.1M is the bridge). No earnings-without-cash red flag.
• Receivables / inventory vs revenue: disciplined — trade receivables $222.8M (≈14% of revenue, +5% YoY on +9% revenue) and inventory actually fell to $256.9M from $268.3M. No channel-stuffing signature.
• Deferred revenue / prepayments: the new and most important line to monitor — customer advances are building (current $25.6M + the $290M Q1'26 SiPho prepayment) and "other long-term assets" jumped to $141M. This is favorable (take-or-pay-like visibility) but introduces revenue-recognition timing judgment — prepayments are credited against future purchases, so the cadence of recognition matters; not a red flag today, but the line to audit each quarter.
• SBC: ~$37.8M FY25 (~2.4% of revenue) flatters non-GAAP modestly; not abusive.
• Goodwill/intangibles: trivial — goodwill $7.0M, intangibles $1.5M. No impairment overhang.
• Tax — a real forward risk: FY25 effective tax ~9% (Israeli 7.5% preferred rate). Management explicitly warns OECD Pillar Two (15% global minimum tax) "would result in significant additional income tax expenses for 2026 and beyond, mainly Israeli operations". A structural margin headwind the bulls' EPS bridge must absorb.
• NCI: TPSCo (51%-owned) runs small losses (NCI −$1.7M FY25); the Mar-2026 Japan restructuring unwinds much of this complexity by Apr-2027.

Regulatory findings (required sub-section):

• SEC Litigation Releases / AAERs: none. regulatory/regulatory-findings.md (generated 2026-06-29 via EDGAR EFTS LR + AAER) returns 0 SEC findings for Tower in 2021–2026.
• Non-SEC enforcement (web): no FTC/DOJ/FDA/CFPB action found. The material legal items are commercial litigation, not enforcement: (1) GlobalFoundries patent suits at the US ITC + W.D. Texas, 11 patents, seeking an import injunction (see Lens 13); (2) the Fab 3 Newport Beach lease dispute — the landlord alleges a "material non-curable breach" over noise-abatement and seeks a judicial declaration; a third party also asserts collateral rights to the site (landlord has agreed to indemnify Tower); (3) the Intel capacity-corridor mediation. The IQE IP dispute was settled in 2026 via a cross-license.
• 10-K/20-F Item 3 (own disclosure): Tower discloses it "is a party to litigation incidental to the conduct of our ongoing business, including class actions, disputes with customers, suppliers" but nothing it deems individually material beyond the items above.
• Net: No accounting or regulatory-enforcement red flags — verified via SEC EDGAR EFTS (LR, AAER = 0), web search, and 20-F Item 3, as of 2026-06-29. The risk surface is IP litigation and forward tax, not fraud or enforcement.

Phase D — Project & stress-test

Lens 11 · Forward Projection

Bottom-up from FY2025 actuals + Q1'26 + guidance + management's own 2028 model. All outputs ``, inputs labeled. No forecast.ts create (watchlist mode). Share count ~114M diluted.

Anchors: FY25 actual rev $1,566M, EPS $1.94 dil. Q1'26 $414M; Q2'26 guide $455M. Management 2028 model: $2,800M revenue / $750M net profit. FY26 Street consensus EPS ~$2.98 (raised from $2.61).

Read: the 2028 bull case (~$6.30 EPS) on today's $250 price is ~40x 2028 earnings — i.e., even if management hits its model, the stock is not cheap two years out; the base case ($5.65) is ~44x. The contracted $1.3B SiPho 2027 revenue is what makes the forward unusually underwriteable (most foundries can't show you 18-month-forward booked dollars). The swing variable is net margin — Tower has never sustained 25%+; getting there requires SiPho to be both high-ASP and high-utilization while Pillar-Two raises the tax rate. Brier forecast (logged conceptually, not written): "TSEM FY2026 non-GAAP EPS ≥ $3.00, p≈0.60, resolves 2026-12-31."

Lens 12 · Bull vs Bear

Bull case. Tower is the merchant-foundry toll-booth on AI optical interconnect at the exact moment the datacenter moves from copper to light. Silicon photonics grew 3x YoY (Q1'26), customers are pre-paying to reserve capacity ($290M in one quarter) and have contracted $1.3B for 2027, and NVIDIA is a named co-development partner for 1.6T modules. The moat is real (analog/SiPho process-lock; few qualified merchant SiPho fabs; supply < demand through 2027). A fortress balance sheet ($990M net cash) funds the $920M expansion without dilution or debt. Operating leverage is visibly inflecting (op-profit +96% YoY). A credible 20-year operator is reinvesting at high incremental ROIC. Earnings surprise vector: every quarter of SiPho beats compounds the 2028 model toward $6+ EPS. The second leg — 300mm 65nm BCD power for AI/datacenter power delivery + automotive — is a free option the market isn't fully paying for.

Bear case (permanent-impairment risks).

1. The multiple is the risk. ~66–74x forward EPS, ~50x EV/EBITDA, on a business whose FY25 ROE was ~7.6% and which has never sustained 25% net margins. The stock discounts the 2028 bull case today; a single soft SiPho quarter re-rates it hard (it has done a ~5x round-trip on sentiment before — Intel 2022→2023).
2. AI-optics is a capex-cycle, and Tower is adding capacity into it. If hyperscaler optical demand digests/overbuilds (as memory/foundry cycles always eventually do), Tower is left with underutilized 300mm SiPho capacity and impairment/under-absorption charges — management explicitly warns AI wafer pricing "may be volatile and difficult to forecast".
3. GlobalFoundries' ITC patent suit (Lens 13) — an injunction blocking import/sale of allegedly-infringing specialty products in the US is a low-probability/high-severity tail that strikes at the SiPho/RF franchise.
4. Pillar-Two tax structurally lifts the tax rate from ~9% toward 15%, shaving the very margin expansion the bull case needs.

Pre-mortem (18 months out, thesis broke): SiPho 2027 contracts slipped or were renegotiated as hyperscalers paused an optics build-out; a second supplier (TSMC's COUPE, GF) took share; the GF ITC case produced an exclusion order or a costly settlement; Pillar-Two + a weaker analog base compressed margins; the stock de-rated from ~70x to ~25x on a still-respectable but no-longer-hyper-growth EPS — a 50%+ drawdown on multiple compression alone.

Are multiples too high? For the base case, yes — you are paying ~40–44x 2028 earnings if management executes flawlessly. The valuation only "works" on the bull path plus continued AI-optics scarcity.

Contrarian view (what the market refuses to see): the market is treating Tower as a SiPho pure-play, but ~80%+ of revenue is still legacy specialty analog/RF/CIS — a slower-growing, cyclically-exposed base. The bull narrative is a thin (if fast-growing) slice riding on top of an ordinary foundry. Conversely, the under-appreciated bull point is the BCD power-management franchise for AI power delivery — a second secular AI-leverage vector that isn't in most "it's a photonics story" models.

Lens 13 · Devil's Advocate (short-seller)

• What structurally breaks the model: AI optical-interconnect demand is early, lumpy, and customer-concentrated. The whole re-rate rests on a handful of hyperscaler/optical-module programs (NVIDIA-aligned). If two programs push out a year, the $1.3B "2027 contracted" number gets renegotiated (these are capacity reservations, not unconditional take-or-pay in perpetuity) and the growth algorithm inverts.
• Concentration: one customer is 11% of revenue; the SiPho upside is even more concentrated in a few optics buyers. A single design-loss or a hyperscaler in-sourcing photonics is a step-down.
• The moat may be weaker than bulls think: GlobalFoundries is suing Tower over 11 patents at the ITC + W.D. Texas, alleging Tower built its specialty processes on GF's R&D and seeking an import injunction. A competitor weaponizing IP against the exact specialty/RF processes that are Tower's crown jewels is the most dangerous, most under-priced overhang. Even a settlement = a royalty drag on the high-margin lines.
• Most dangerous competitor bulls underestimate: TSMC — Broadcom's CPO (COUPE) optical engines run on TSMC, not Tower. If TSMC decides merchant SiPho at scale is strategic, Tower's scarcity moat erodes fast. GF (larger, 300mm, and litigious) is the second threat.
• Worst capital-allocation risk: building $920M of SiPho/300mm capacity into a possible AI-optics top, funded by customer prepayments that must be recognized as delivered — if demand fades, Tower is long stranded capacity and short the obligations.
• Assumptions that must hold for today's price: (1) SiPho compounds ~3x→steady-high-growth through 2028; (2) net margin reaches ~25% (never achieved); (3) GF litigation is immaterial; (4) Pillar-Two is absorbable; (5) AI-optics demand > supply persists. All five.
• If growth disappoints 20–30%: 2028 revenue ~$2.0–2.2B and net profit ~$450–500M (≈$4 EPS); at a de-rated ~25x that's a ~$100 stock — ~60% downside from ~$250.
• Single scenario that permanently impairs: an ITC exclusion order on Tower's RF/SiPho specialty products (or a forced costly redesign), coincident with an AI-optics digestion — simultaneously hitting the moat and the demand. Plausibility: low-to-moderate, but it is the asymmetric tail the multiple ignores.

Lens 14 · Management Questions (ordered by information value)

1. The $1.3B of 2027 SiPho revenue — what exactly is contracted? Are these unconditional take-or-pay commitments or capacity reservations the customer can renegotiate/walk from, and what are the cancellation terms and prepayment-forfeiture mechanics?
2. GlobalFoundries' ITC + W.D. Texas suit over 11 patents seeking an import injunction — what is your exposure on the SiPho/RF processes specifically, your timeline to a determination, and have you reserved or modeled a settlement/royalty?
3. Your 2028 model is $2.8B revenue at $750M net profit — a ~27% net margin you have never sustained. Bridge it: how much is SiPho ASP, how much utilization/op-leverage, and how much does Pillar-Two subtract?
4. How concentrated is the SiPho book by end-customer, and what is your single-customer revenue concentration risk if one hyperscaler in-sources or delays its optics roadmap?
5. Pillar Two — quantify the FY26 and FY27 cash-tax increase versus the historical ~9% effective rate, and what mitigations (IP location, incentives) are available?
6. With ~$990M net cash and no dividend/buyback, why reinvest 100% into capacity rather than return capital — and what incremental ROIC are you underwriting on the $920M program?
7. TSMC supplies Broadcom's CPO optical engines; what is your defensible advantage as a merchant SiPho foundry if TSMC scales merchant photonics, and where do you win on PDK/maturity/capacity?
8. The Intel New Mexico 300mm capacity-corridor — assume it's gone: does losing that 300mm optionality constrain your SiPho/power ramp, and what is the back-up capacity path?
9. The Japan restructuring (Tower takes 300mm Fab 7, Nuvoton takes 200mm Fab 5, closing Apr-2027) — what is the net capacity, cost, and margin effect, and does it create any supply gap for current customers?
10. The Agrate 300mm fab is managed by ST and you own ~1/3 of the cleanroom — what governance/expansion rights do you have if you need more 300mm capacity than ST will allocate?
11. Silicon-photonics gross margins vs the corporate average — are these structurally higher-margin wafers, and where does blended gross margin settle at the 2028 revenue level?
12. Financing income was ~24% of FY25 pre-tax profit; as you spend the cash on capex and if rates fall, how much of that earnings cushion goes away?
13. The Fab 3 Newport Beach lease — what is the realistic worst case on the landlord's "material breach" claim and the third-party collateral assertion, and is relocation of that SiPho capacity ever on the table?
14. Capacity-reservation prepayments are credited against future purchases — walk through the revenue-recognition timing and how investors should model deferred-revenue draw-down quarter to quarter.
15. Beyond photonics: how big can the 300mm 65nm BCD power-management franchise (AI/datacenter power delivery + automotive) become by 2028, and is it in the $2.8B target or upside to it?